Background
Model-informed precision dosing (MIPD) requires a sufficiently predictive model. Both the model reported by Thomson1 and by Goti2 have been validated as fit-for-purpose for dosing vancomycin in adult patients with a BMI < 40 kg/m2 3,4. Recently, modifications have been made to enhance their predictiveness. The modified Goti model5, where serum creatinine rounding to 1 mg/dL for adults over 65 years old was removed from the original model, was found to be more predictive than the published model. Similarly, the Thomson model with a creatinine clearance cap of 150 mL/min ("capped Thomson") was found to be more predictive than the published model6. While both models showed good ability to predict vancomycin levels in retrospective analyses4,6, their relative ability to guide MIPD prospectively was unknown. InsightRX Nova is model-agnostic, and partner sites have varied in which model they select to guide dose selection in this patient population, allowing assessment of model choice on drug exposure.
Methods
Patient data collected during routine clinical care of adult patients treated with intravenous vancomycin and entered into InsightRX Nova between January 1st 2020 and June 15th 2023 were de-identified and analyzed retrospectively. Patients were included if they were at least 18 years old, had a BMI < 40 kg/m2 and had at least one vancomycin level collected, resulting in a data set of 168,000 treatment courses across 117 hospital systems.
Patients were split into groups dosed with Goti models (published Goti, modified Goti) and those dosed with Thomson models (published Thomson, capped Thomson). For each patient, all available drug concentrations were utilized to inform maximum a posteriori (MAP) Bayesian estimates of individual PK parameters. These parameters were then used to estimate average AUC24 across the entire treatment course. AUC24 was estimated using the modified Goti model and the capped Thomson model, and it was considered to be on-target if it fell between 400-600 mg-h/L.
For each patient, population PK parameters were used to predict the value of the first serum vancomycin level. Subsequent levels were predicted using individualized PK parameter estimates through MAP Bayesian estimation, and these parameters were iteratively used to predict the next level. Prediction error was assessed using accuracy, root mean square error (RMSE), and mean percent error (MPE) of the predictions. A prediction was deemed accurate if the prediction was either within 15% or within 2.5 mg/dL of the observed level7.
Results
We found that the Thomson-dosed group had higher AUC24 target attainment rates than the Goti-dosed group. In the figure below, AUCs are estimated by the modified Goti and capped Thomson model for every patient dosed with the Goti (top row) or Thomson (bottom row) models. The number needed to treat using the Thomson model to get one more patient on AUC target than the Goti model is 11 for both estimation models.
The use of the Thomson model remained statistically significantly correlated with higher target attainment even after controlling for other factors, such as age, renal impairment, and BMI (data available upon request).
In terms of prediction error, the Thomson models had higher accuracy, whereas the Goti models had slightly lower RMSE and a posteriori bias. However, the Goti models over-predict levels a priori, which, when combined with a likely clinical tendency to dose conservatively and the time needed to accumulate drug concentrations, leads to initial underdosing in patients and therefore lower AUC24 target attainment in patients dosed with Goti models.
Conclusion
As of the date of this publication, InsightRX recommends using the capped Thomson model for MIPD of vancomycin in adult patients with BMI less than 40 kg/m2.
Manuscript to be published by InsightRX. Data on file and available at reqeust.
1. Thomson et al. J. Antimicrob Chemother. (2009).
2. Goti et al. TDM. (2018).
3. Broeker et al. Clin Microbiol Infect. (2019).
4. InsightRX. Adult vancomycin model comparison for not extremely obese patients. (2020).
5. Tong et al. TDM. (2021).
6. InsightRX. Capped CRCL for adult vancomycin models. (2022).
7. Hughes and Keizer. CPT Pharmacometrics Syst Pharmacol. (2021).
